Microsoft Sets Sights On Polling Firms

Microsoft is using the wisdom of crowds theory to run a prediction service. It may use the technology to turn its Siri-alternative into an opinion pollster.

After some previous trials, Microsoft has now launched the Microsoft Prediction Lab. It’s similar to an opinion poll or survey but rather than ask people what they want to happen (such as the winner of an election) it asks what they think will happen.

That’s not a new idea: it’s based on a theory that’s described in a book by James Suriowecki. The basic idea is that if you ask enough people to make a prediction and average the results, you’ll pool together all sorts of individual knowledge and insights. A common example given of the theory is that in fair games such as “guess the weight” or “guess the number of jelly beans in the jar” the average of all the guesses will often prove closer to the true figure than any individual guess.

Microsoft Prediction Lab turns that process into a sort of game by allowing players to stake points on their predictions. That gives an added factor of being able to effectively measure how confident each prediction is. The point system works just like a betting market: the “return” for each point staked is based on how likely or unlikely your prediction seems based on those who’ve already made a prediction.

The first version of the game will be based on politics and will run until US election day in November. For now it appears it’s for fun only, with no prizes.

In the long term, Microsoft plans to collect predictions or opinions via Cortana, its mobile assistant equivalent to Apple’s Siri. The idea will be to use the digital voice in place of having a pollster phone a member of the public at random. That could make it easier to “poll” the same people at regular intervals and track whether specific opinions or predictions change over time or in response to specific events.

While there are serious data protection issues to address, Microsoft could theoretically do a better job of putting together a panel of respondents that specifically matches the demographics of a particular group, such as voters in a state or Congressional district.